AI Won't Reduce Your Work. It Will Multiply It (Until It Doesn't)

The steam engine did not reduce coal consumption. AI will not reduce your workload. Until it does, all at once, or?

in 1865, economist Willian Stanley Jevons observed something interesting. as steam engines became more efficient at burning coal, the total coal consumption increased, not decreased as expected. Better efficiency made coal-powered machines so economical that demand exploded.

The same phenomenon we see with AI in knowledge work. And as I believe very soon in physical AI, aka in robots, that might lead also the “steam engine revolution 2.0”.

But let us take some steps back.

AI Productivity Treadmill

Imagine a world where AI-assisted developers or fully AI developed artifacts take hours instead of weeks or month. Your customers, internal as external, will expect that you provide now 10x features for the cost and time of one. At the same we hear already expectations that prices should decrease for that work. I believe this costs first will accelerate more, token prices tend to rise at the moment. We are practically still in the beta testing of steam, engines. Its a mass test, but its not optimized. (also we live very much in a bubble).

However, I think this direction will lead to mountains of delivered work, that no one every will look at. And in worst case, is just consumed by other AIs. We will have more output, but not less work.

And it happens already. Companies using AI-powered coding tools report that while individual tasks get done faster, the volume of tasks explodes but the cycle of useful releases decrease.

The best study I know on this is a randomized trial by METR from July 2025: 16 experienced open-source developers (I know, scientifically not super representative), 246 real tasks on their own code bases, with and without Cursor. With AI they were 19% slower, but hey believed they were 20% faster.
Companies using AI-powered coding tools report the same pattern. Individual tasks feel faster, the volume of tasks explodes, but the cycle of useful releases slows down. Google’s DORA data puts a number on it: a 25% increase in AI adoption correlated with 7.2% less delivery stability. More code, more ...

The hidden costs of cognitive load, quality and coordination

I’m coming from a background that we call nowadays Platform Engineering. Think of it like we are the construction workers who built the basis, all required connections, but also all tools that are needed to built something, and afterwards operate it, and keep developing it. I know, abstract, and in addition we do that in a continuous moving environment where tools get upgraded every 3-4 month. Due to this complexity and speed of development, the term cognitive load became very prominent.

And the same issues with raised cognitive load will fritzlblitz the average Jane/Joe in their white collar job. Professionals will, and have to already, deal with tons of AI tools, outputs and so on. This requires a structure coordination and control where happens what. It requires a fast human judgement and a high level of discipline to keep a good level of quality.

Frankly speaking, I’m using AI for a lot of stuff in work, for projects, for private things, to do research and to challenge me. And I have days were it is just to much of information. I don’t think I’m alone.

This all shifts our costs when we’re producing more while understanding less.

• From creation to comprehension (writing code is easy; maintaining it is hard).

• From building to maintaining (more features mean more and faster legacy systems).

• From output to oversight (someone still has to ensure quality).

Leadership will be challenged hard

In technical leadership, the temptation will be to measure success by velocity alone. “how much can we ship, how fast can we move?” But if your team’s output increases 10x, your investment in code review, testing infrastructure and architectural oversight must scale proportionally or you’re just accelerating toward chaos. This is valid also to other domains, but I use here as an example things I’m very familiar with.

I want to cite here the University of Chicago economist Alex Imas and Apollo Global Management’s chief economist Torsten Slok: “If you automate 90% of the job, then everyone does the 10% of the job. And the 10% kind of expands to be 100% of what people do and kind of 10xs their productivity.” (source: https://fortune.com/2026/05/05/dario-amodei-jevons-paradox-will-ai-wipe-out-white-collar-jobs/)

We will see a very drastic acceleration in this behavior. And the market will be self propelling. The same as AI research companies are locked into competition of who has build fast the best model or reach AGI, because stakeholder wants to see it, because people should spend money on you.

That means for us, as person, doesn’t matter at which desk job, we have to become more and more efficient in using tons of AI. For some time this might mean that we need more people to oversee outputs. And we don’t know how long this goes on. Interestingly enough so, token costs are also rising. So This increase comes with costs too, and they are not little.

Early adopters leadership teams have learned this already the hard way. (also I don’t understand how they came to that ideas at that time, when looking back how poorly performing AI models where 1-2 years ago...). Klarna replaced roughly 700 support agents with an AI assistant and celebrated the savings. In 2025 the CEO admitted: “We went too far. We focused too much on cost. The result was lower quality.” (You don’t say Sherlock). They are hiring humans again. Duolingo made AI usage part of performance reviews in 2025 and dropped it again in 2026, after employees asked whether they were supposed to use AI for AI’s sake. Something many people right now are facing. Use AI for the sake of AI.

You can observe this in IT very well, because we do a lot publicly visible, which means the part thats not public is even more drastic in its behavior. Looking at Open Source and GitHub, from 2024 to 2025:

• number of commits increased by 25%

• merged pull requests increased by 23%

• 230 new repositories are created every minute

• and GitHubs Copilot had authored alone over 1 Million PRs in just 5 months in 2025...

• AND comments on commits has dropped by 27%
  I don’t want to know how this looks like for 2025-2026.

One day, it all will flip

But there will be one point where, practically from one day to another, companies don’t need the majority of their staff anymore.

It is a long way until then. Agents, AI, need to learn and understand companies inside out. At some point they optimize all the false processes we human had build. It will tear down kingdoms, it will make all oversight obsolete.

I know know, it sounds like a dystopia. I think that just depends on your job.

Stanford’s Brynjolfsson tracked 25 million workers through ADP (that’s an HR tool) payroll data. Employment of 22 to 25 year old in the most AI-exposed jobs is down 19% by mid-2026, while experienced workers are untouched so far. Companies are not firing seniors. They are simply not hiring juniors. (That’s actually an issue I also have. For my teams and projects we need Seniors, real Seniors, that’s what custoemr asks for). Big Tech new-grad hiring is down about 50% from pre-pandemic. At the same time, AI-attributed layoffs are still under 1% of all US job cuts in 2025.

So the collapse is not here. But the pipeline that produces the next senior engineers, lawyers and consultants is quietly being switched off.

You might call the people crazy who are very close to this research and leaving their jobs because they are afraid. It is maybe 5 years out, maybe 20 in Europe :D Maybe just 2 or less...

Until then our jobs will change a lot the next years. You, me, anyone should work on their flexibility, you might become really fast obsolete. (And I’m fully there with you, the AI madness, is just that, pure madness. But it is inevitable.)

What’s your take on it? As leader? As employee? As one working more in a blue collar work?